Frontier models waver on decision theory depending on who asks
LessWrong discussions reveal frontier models shift their stated decision-theory positions based on the asker's implied background, with sycophancy rates estimated at 30-100%, while a new essay argues against strong decision-theoretic realism.
2026-10-01 ~ 2026-10-01 · 3 related posts
- LessWrong essay argues against strong decision-theoretic realism, highlighting ASI path dependence — dfrsrchtwts · 2026-10-01
- GPT-6 Astra prefers CDT over EDT, unlike Anthropic models, suggests decision-theory sycophancy gap — dfrsrchtwts · 2026-10-01
- Frontier models state different decision theory preferences depending on who's asking, per LessWrong analysis — dfrsrchtwts · 2026-10-01